JP5998779B2 - SEARCH DEVICE, SEARCH METHOD, AND PROGRAM - Google Patents

Description

  The embodiments discussed herein relate to information retrieval techniques.

  As one of information retrieval techniques, a technique called fuzzy retrieval is known (see, for example, Non-Patent Document 1, Non-Patent Document 2, and Non-Patent Document 3). The fuzzy search is a search technique that makes a matching succeed even when a search target text does not completely match a specified search character string pattern, and is widely used in a text search system.

  For example, an ambiguous search is performed with respect to a text database storing millions of sentence examples using a search keyword “I want to take a picture with an easy phone”. Then, for example, search results such as “How to take a photo with easy phone”, “Precautions when taking a photo”, “When you want to send a photo by email”, etc. It is done. As described above, in the fuzzy search, the search character string such as the search result “Raku Raku Fon” for the search character string “Raku Raku Fon” or the search result “Shoot” and “Shoot” for the search character string “I want to shoot” is used. Text that does not match exactly is also obtained as a search result.

  Also known is a machine translation technique in which translation from an original sentence in a first language to be translated into a sentence in a second language is performed by a machine such as a computer. One of the technologies is to display the difference between the translation source text and the translation example source text in correspondence with the words in the translation example translation, distinguishing it from other words, and the equivalent of the translation example translation difference. A technique for facilitating editing is known (for example, see Patent Document 1). Also known is a technique for improving translation accuracy by providing a machine translation device with a function for automatically acquiring knowledge for translation so that a document search device in the first language can be used in the second language. (For example, refer to Patent Document 2). Furthermore, a range related to the information to be obtained is specified for document information in a certain language, and an information search formula for searching for a document similar to the document in the specified range is generated based on the machine translation result of the document in the specified range. A technique of searching for documents in other languages is known (see, for example, Patent Document 3).

JP 2009-116584 A JP-A-11-203287 JP 2005-11138 A

Gene Myers, “A fast bit-vector algorithm for approximate string matching based on dynamic programming) ", Journal of the ACM, Association for Computing Machinery, May 1999, 46, 3, p.395- 415 Gonzalo Navarro, "A guided tour to approximate string matching", ACM Computing Surveys, Association For・ Association for Computing Machinery, March 2001, Vol.33, No.1, p.31-88 Gonzalo Navarro, 3 others, "Indexing Methods for Approximate String Matching", Bulletin of the Technical Committee on Data Engineering (Bulletin of the Technical Committee on Data Engineering), IEEE Computer Society, December 2001, Vol. 24, No. 4, p.19-27

The above-described fuzzy search technique is merely a matching of character faces, and therefore, there is a known problem that it is vulnerable to word-level notation fluctuations.
For example, consider a case in which an ambiguous search is performed using a query sentence “apple varieties popular in Japan”. In this query statement, “apple” written in katakana is widely used as “apple” written in hiragana and “apple” written in kanji. In addition, the character string “popular” in this query statement is also widely used as a synonym “popular”. In the fuzzy search, such “apple”, “apple”, and “apple” matching, or “popular” and “popular” matching cannot simply be performed on sentence examples in a text database. In order to enable such matching, for example, processing such as expanding a search key before searching using a synonym dictionary is performed.

In addition, when the fuzzy search technique is applied to a language in which the word order is relatively free, such as Japanese or Korean, the search accuracy may be reduced due to fluctuation of the word order.
For example, let us consider the case where an ambiguous search is performed using a query sentence “a variety of apples popular in Japan”. In this case, it is impossible to determine that the meanings of the sentences in which the word order is changed are the same unless the search is simply based on whether the character strings are similar. For this reason, for example, even if a text example such as “Japanese apple varieties are popular in Japan” exists in the text database, the similarity of the query sentence is low, and the rank of the search result falls below. Sometimes. That is, in this case, for example, sentence examples such as “San Fuji is the most popular apple variety in Japan” may be arranged in the lower order of the search results. And in this case, for example, “Apple apple pie is popular in Japan”, “Japanese are particular about the type of apple”, “Three puddings that are popular in Japan” Sentence examples such as "" are arranged at the top of the search results. Furthermore, in the case where the above-mentioned word level notation is not taken into account, for example, “The most popular types of apples are Fuji, Tsugaru, Jonagold, etc. in Japan.” The sentence example that should be the highest search result may sink to the lower level.

  In view of the above-described problems, a search device described later in this specification suppresses a decrease in search accuracy of an ambiguous search caused by word notation and fluctuation of word order widely accepted depending on a language to be used.

One of the search devices described later in this specification includes a bilingual format database, a machine translation unit, a determination unit, and an ambiguous search unit. Here, in the bilingual format database, a question type representing a type of a question to which a text in the first language to be searched, a translated sentence in the second language for the text, and a phrase included in the text are answers Information is stored in association with each other. The machine translation unit machine translates the input query sentence in the first language into the second language. The determination unit determines the type of the question expressed by the query sentence. The ambiguous search unit performs an ambiguous search using the query sentence translated into the second language with respect to the translation sentence stored in the parallel translation database and the type of the question determined by the determination unit as a search key. Then, the fuzzy search unit extracts the text of the first language associated with the translated sentence that is the result of the fuzzy search from the bilingual format database and outputs it.

According to the search device described later in this specification, there is an effect that a decrease in search accuracy of fuzzy search caused by word notation and fluctuation of word order widely accepted depending on a language to be used is suppressed.
There is an effect.

It is a functional block diagram of one Example of a search device. It is the figure which illustrated the procedure of the machine translation by a machine translation part. It is the figure which illustrated an example of the conceptual structure of the original sentence produced | generated by the semantic analysis part. It is explanatory drawing of the method of the production | generation of the meta information for original texts. It is the figure which illustrated the data structure of the bilingual form database. It is the figure which illustrated the example of creation of the conceptual structure of a query sentence. It is explanatory drawing of the method of the production | generation of the meta information for query sentences. It is explanatory drawing of creation of the search key by a search key creation part. It is a specific example of the fuzzy search by a fuzzy search part. It is a figure showing the hardware structural example of one Example of a search device. It is a flowchart showing the processing content of the 1st example of bilingual form DB creation processing. It is a flowchart showing the processing content of the 1st example of a search process. It is a flowchart (the 1) showing the processing content of the 2nd example of bilingual form DB creation processing. It is a flowchart (the 2) showing the processing content of the 2nd example of bilingual form DB creation processing. It is a flowchart (the 1) showing the processing content of the 2nd example of a search process. It is the flowchart (the 2) showing the processing content of the 2nd example of a search process. It is a flowchart (the 3) showing the processing content of the 2nd example of a search process.

First, FIG. 1 will be described. FIG. 1 is a functional configuration diagram of an embodiment of a search device 1.
When the query sentence 2 expressed using the first language is input, the search device 1 of FIG. 1 performs an ambiguous search on the database using the query sentence 2 as a search key, A device that outputs text expressed using a first language.

The search device 1 of FIG. 1 includes a source text database 10, a parallel translation format database 11, a search key creation unit 13, and an ambiguous search unit 14.
The original text database 10 stores a large number of texts (hereinafter referred to as “original texts”) expressed using the first language, which are search targets of the fuzzy search performed by the search device 1. Database. In the following description, the original text database 10 is referred to as “original text DB 10”.

  The bilingual format database 11 is a database in which a large number of original sentences stored in the original text DB 10 and translated sentences of the original sentences in the second language are stored in association with each other. In the following description, the parallel translation format database 11 is referred to as a “parallel translation format DB 11”.

  The machine translation unit 12 machine-translates the query sentence 2 input to the search device 1 and expressed using the first language into the second language. In the present embodiment, the translated text stored in the parallel translation format DB 11, that is, the translated text of the original text in the second language is also created by machine translation by the machine translation section 12.

The search key creation unit 13 creates a search key in the fuzzy search performed by the fuzzy search unit 14 from the query sentence 2 machine-translated into the second language by the machine translation unit 12.
The ambiguous search unit 14 performs an ambiguous search using the search key created by the search key creating unit 13 for the translated sentence stored in the parallel translation format DB 11. Then, the original sentence associated with the translated sentence obtained as a result of the fuzzy search is extracted from the bilingual form DB 11 and output as the search result 3.

  In the search device 1 having the above-described configuration, an ambiguous search using the second language translation of the query sentence 2 as a search key is performed on the second language translation of the original sentence to be searched. Is called.

In the following description, a case where the first language is Japanese and the second language is English will be described as an example.
Assume that the query sentence 2 is “Taro ate an apple at school”, and the original sentence “Taro ate an apple at school” is vaguely searched using the query sentence 2 as a search key. In the conventional fuzzy search, character strings are matched between Japanese, so that “apple” in the search key and “apple” in the original text are determined as different words. For this reason, the rankings of these two sentences may drop despite the exact same meaning. However, the English translation of the query sentence 2 “Taro ate an apple at school” and the English translation of the original sentence “Taro ate an apple at school” are both “Taro ate an apple”. at school. "is exactly the same sentence. Accordingly, the ranking of the search result 3 by the search device 1 is the original sentence “Taro ate apple at school”, and the most appropriate result is obtained.

  As another example, query sentence 2 is “Taro ate an apple at school.” Using this query sentence 2 as a search key, the original sentence “Taro ate an apple at school” is ambiguously searched. Assume a case. In the conventional fuzzy search, character strings are matched between Japanese, but the order of “Taroga” and “in school” is different between the query sentence 2 and the original sentence. Despite the same meaning, the ranking may drop. However, the English translation of “Taro ate an apple at school”, which is the query sentence 2, and the English translation of the original “Taro ate an apple at school” are both “Taro ate an apple”. at school. "is exactly the same sentence. Accordingly, the ranking of the search result 3 by the search device 1 is the original sentence “Taro ate an apple at school”, and the most appropriate result is obtained.

  As described above, in the ambiguous search using the search device 1, since the word expression and word order fluctuation in the first language in the original sentence and the query sentence 2 are absorbed, a decrease in search accuracy due to these fluctuations is suppressed. Is done.

The search device 1 in FIG. 1 further includes a query sentence question type determination unit 21 and a query sentence question type determination pattern 22.
The query sentence question type determination unit 21 determines the type of the question expressed by the query sentence 2 by using the query sentence question type determination pattern 22. The query sentence question type determination pattern 22 will be described later.

  Here, in the bilingual translation format DB 11, question type information indicating the type of question for which the word / phrase included in the original text is an answer is further stored in association with the translated text of the original text. Further, the search key creating unit 13 further creates a search key in the fuzzy search performed by the fuzzy search unit 14 from the question type determined by the query sentence question type determining unit 21. Then, the ambiguous search unit 14 performs an ambiguous search using the question type determined by the query sentence question type determining unit 21 in addition to the query sentence 2 translated into the second language as a search key. Then, the original sentence associated with the translated sentence obtained as a result of the fuzzy search is extracted from the bilingual form DB 11 and output as the search result 3.

  Since the search device 1 has the above-described configuration, it is possible to search for an answer to a so-called 5W1H type question. The “5W1H type question” refers to a question asking about the subject (Who), the target (What), the time (When), the place (Where), the reason (Why), or the method (How).

In the conventional fuzzy search, since the search is performed by matching character strings, there is a case where the search for an answer to such a 5W1H type question cannot be supported.
For example, when it is desired to know when to plant rice in Sendai, the query sentence 2 is simply “When do you plant rice in Sendai?” And the original sentence is searched using this query sentence 2 as a search key. . In the conventional fuzzy search, in this case, “When to plant rice in Sendai is a problem.” “It is necessary to investigate when to plant rice in Sendai.” “When will rice be planted in Sendai?” The original text that includes the character string “when” appears at the top of the ranking. On the other hand, in the conventional fuzzy search, in this case, the original information including the character string “June”, for example, the original text “Sendai planting rice in June.” Rankings may sink below.

  On the other hand, in the search device 1 having the above-described configuration, the query sentence translated into the second language with respect to the translation sentence associated with the information indicating the type of the question whose answer is the original phrase. An ambiguous search is performed using 2 and the type of question of the query sentence 2 as search keys. In other words, in the above example, the original sentence “The rice plant is planted in Sendai in June.” In the original sentence “The rice plant is planted in Sendai in June.” Information “When” indicating the type is associated with and stored in the parallel translation format DB 11. Further, when the query sentence 2 is “When do you plant rice in Sendai?”, The query sentence question type determination unit 21 determines that the type of the question expressed by the query sentence 2 is “When”. Determined. As a result, an ambiguous search is performed using the English translation of the query sentence 2 “When is the rice planted in Sendai?” And the question type “When” expressed by the query sentence 2 as search keys. Therefore, in addition to the English translation sentence of the query sentence 2 and the English translation sentence of the original sentence being similar, it can be expected that the original sentence in which both types of questions match appears in the higher rank of the search result 3. That is, the original text including the information to be obtained is obtained as the search result 3 with higher accuracy.

The search device 1 of FIG. 1 includes a semantic analysis unit 31 and a meta information generation unit 32.
The semantic analysis unit 31 performs semantic analysis of the original text stored in the original text DB 10.
In the configuration of FIG. 1, the semantic analysis function provided in the machine translation unit 12 is used as the semantic analysis unit 31. That is, in the configuration of FIG. 1, the machine translation unit 12 generates a sentence in the second language corresponding to the result of the semantic analysis by the semantic analysis unit 31 for the sentence to be translated expressed in the first language. Then, machine translation of the sentence to be translated into the second language is performed. Here, the semantic analysis unit 31 may be provided in the search device 1 separately from the machine translation unit 12.

The meta information generation unit 32 includes an original text question type determination unit 33 and a storage processing unit 35.
Based on the result of the semantic analysis of the original text by the semantic analysis unit 31 and using the original text question type determination pattern 34, the original text question type determination unit 33 uses the original text question type determination pattern 34 to determine the phrase included in the translation of the original text. Generate question type information. The original question type determination pattern 34 will be described later.

The storage processing unit 35 stores the question type information generated by the original text question type discriminating unit 33 in the parallel translation format DB 11 in association with the translated text of the original text.
The semantic analysis unit 31 and the meta information generation unit 32 store the question type information about the words included in the original text in the parallel translation format DB 11. The ambiguous search unit 14 performs an ambiguous search on the translated sentence of the original sentence and the question type information associated with the translated sentence stored in the parallel translation format DB 11 in this way.

  In the present embodiment, the source text question type discriminating unit 33 includes the above question type information and a phrase included in the translation of the original text that is an answer to the question of the type represented by the question type information. Generate meta information. Then, the storage processing unit 35 adds the meta information generated by the original text question type determination unit 33 to the translated text of the original text and stores it in the parallel translation format DB 11.

  For example, assume that the original text is “Plant to plant rice in Sendai in June”. When the semantic analysis unit 31 performs semantic analysis on the original text, a conceptual structure of the original text is obtained. Based on this conceptual structure, first of all, “In Sendai, we plant. It can be seen that the portion of "contains the location information about planting. In this case, the text question type discriminating unit 33 uses meta-information “@Where PLANT” in which the translated word “PLANT” of “planting” is used as a central independent word and information “@Where” indicating that it includes location information is combined. Is generated. Moreover, based on the above-mentioned conceptual structure obtained by the semantic analysis unit 31, “plant in June. It can be seen that the part of "" also contains timing information about planting. In this case, the text question type discriminating unit 33 uses meta-information “@When PLANT” in which the translated word “PLANT” of “planting” is used as a central independent word and information “@When” indicating that it includes time information is combined. Also generate. The storage processing unit 35 adds “@Where PLANT @When PLANT The rice plant is planted in Sendai in June.” With the generated meta information “@Where PLANT” and “@When PLANT” added to the English translation of the original text. It is stored in the bilingual format DB 11. The ambiguous search unit 14 performs an ambiguous search on the translation sentence added with the meta information stored in the parallel translation format DB 11 in this way.

  In the present embodiment, the semantic analysis unit 31 further performs a semantic analysis of the query sentence 2 input to the search device 1. Therefore, in the configuration of FIG. 1 in which the semantic analysis function of the machine translation unit 12 is used as the semantic analysis unit 31, the machine translation unit 12 uses the second language corresponding to the result of the semantic analysis of the query sentence 2 by the semantic analysis unit 31. Machine translation of the query sentence 2 into the second language. The query sentence question type determination unit 21 is expressed by the query sentence 2 based on the result of the semantic analysis of the query sentence 2 by the semantic analysis unit 31 and using the query sentence question type determination pattern 22. Determine the type of question you have.

  For example, it is assumed that a query sentence 2 “When do you plant rice in Sendai?” Is input to the search device 1. When the semantic analysis unit 31 performs semantic analysis on the query sentence 2, the conceptual structure of the query sentence 2 is obtained. Here, the query sentence question type determination unit 21 determines which question type of the query sentence 2 is a 5W1H type question by paying particular attention to the question words. In this example, "When do you plant?" From the part, it turns out that it is a question about time information about planting. In this case, the query sentence question type discriminating unit 21 uses the translated word “PLANT” of “planting” as a central independent word and meta information combining information “@When” indicating that the question type is a question of time information. Generate “@When PLANT”. The search key creation unit 13 creates "@When PLANT When is the rice plant planted in Sendai?" By adding the generated meta information to the English translation of the query sentence 2 translated by the machine translation unit 12 To do. The ambiguous search unit 14 performs an ambiguous search on the translation of the original sentence stored in the parallel translation format DB 11 using the search key created in this way.

  In this way, the search device 1 performs an ambiguous search on the bilingual form DB 11 using the meta information created as described above and the translation of the query sentence 2 as search keys. Accordingly, in addition to the fact that the translation sentences of the query sentence 2 and the original sentence are similar to each other, the original sentence in which the meta information added to both is matched can be obtained as the higher rank of the search result 3. I can expect. Therefore, the original text including the information to be obtained is obtained as the search result 3 with higher accuracy.

The search device 1 in FIG. 1 is configured as described above.
Next, the operation of generating meta information and storing it in the bilingual form DB 11 by the meta information generating unit 32 will be described in more detail.

First, FIG. 2 will be described. FIG. 2 illustrates a machine translation procedure performed by the machine translation unit 12.
In the machine translation by the machine translation unit 12, each process of S1: morphological analysis, S2: syntax analysis, S3: semantic analysis, and S4: sentence generation is sequentially performed. The machine translation method itself is a widely known method.

  The morpheme analysis process of S1 is a process of dividing the original text to be translated expressed in the first language into morphemes and determining the part of speech of each morpheme. In this morphological analysis process, a translation dictionary 40, which is a list in which words used in the original sentence, parts of speech of the words, and translations of the words in the second language are associated with each other, is used. In the configuration of FIG. 1, the translation dictionary 40 is provided in the machine translation unit 12 itself.

  The syntax analysis process of S2 uses the morpheme obtained by the morpheme analysis process and the discrimination result of the part of speech to analyze the syntactic structure of the original sentence, thereby providing a grammatical relationship between the words constituting the original sentence ( Create a syntax tree representing the dependency structure between clauses).

  The semantic analysis process of S3 generates a conceptual structure of the original sentence by performing a semantic analysis on the original sentence syntax tree obtained by the syntactic analysis process. In the present embodiment, the conceptual structure of the original text includes a node (node) representing the word constituting the original text and an arc (arc) representing the conceptual relationship between the words by connecting each node. It is expressed by the tree structure used.

  In the sentence generation process of S4, candidate translations corresponding to the words constituting the original sentence are extracted from the translation dictionary 40, and a translation word is selected from the extracted candidates based on the conceptual structure generated by the semantic analysis process. Is a process of generating a translated sentence in the second language for the original sentence.

  For example, in FIG. 2, the original text “Taro bought a Fujitsu computer and mobile phone in Akihabara” was translated into “Taro bought the personal computer and the cellular phone of Fujitsu in Akihabara.” The procedure is illustrated.

  The semantic analysis unit 31 in the search device 1 of FIG. 1 performs the processes of S1: morphological analysis, S2: syntax analysis, and S3: semantic analysis in the machine translation procedure by the machine translation unit 12 described above.

Here, FIG. 3 will be described. FIG. 3 illustrates an example of the conceptual structure of the original text generated by the semantic analysis unit 31.
The example in Fig. 3 shows the sentence in the first paragraph of the sentence "The Chao Phra River, which flows down the central part of Thailand, is the first river in Thailand, and the Chao Phraya Delta in the downstream area is a major rice producing region in the Asian monsoon region. (FIG. 1) ”is a conceptual structure created by the semantic analysis unit 31.

  The question type discriminating unit for original text 33 of the meta information generating unit 32 is the result of the semantic analysis on the arc in the conceptual structure of the original text that matches the predetermined pattern previously associated with the question type. Meta information is generated. The meta information is generated by using a phrase represented by nodes connected by arcs and a question type. A method of generating the meta information will be described with reference to FIG.

  The conceptual structure illustrated in FIG. 4 is the same as that in the example of FIG. Further, all of the extraction patterns a, b, c,..., N illustrated in FIG. 4 are the original text question type determination pattern 34 in FIG. The original question type discriminating pattern 34 is a pattern in which the result of the semantic analysis on the arc in the conceptual structure of the original sentence is associated with the question type.

  For example, “extraction pattern a” in FIG. 4 represents a question that asks for a place among the types of 5W1H type questions when the result of the semantic analysis of the arc in the conceptual structure of the original text is “place”. It indicates that “Where” is associated. That is, this extraction pattern indicates that meta information “@Where” is added to the translation of the original text when such an arc exists in the conceptual structure of the original text.

  Further, for example, “extraction pattern c” in FIG. 4 represents a question asking the target among the types of 5W1H type questions when the result of the semantic analysis on the arc in the conceptual structure of the original text is “target”. "What" is associated. That is, this extraction pattern indicates that meta information “@What” is added to the translation of the original text when such an arc exists in the conceptual structure of the original text.

  In this extraction pattern, if such an arc exists in the conceptual structure of the original text, the translation of the phrase represented by the node on the root side of the arrow representing this arc is included in the meta information as a central independent word. This also indicates that it is added to the translation of the original text. By including this central independent word in the translation information and adding it to the translated sentence, not only the 5W1H type of question but also a sentence that matches not only the question type but also the related word is ambiguous by the ambiguous search unit 14. In the search result 3 obtained by the search, it can be expected that the rank is further increased. The central independent word may not be included in the meta information.

  In the example of FIG. 4, the arcs (A) and (B) are detected as part of the arc whose semantic analysis result is “target” from the conceptual structure, and the arc whose semantic analysis result is “location” is detected. It is assumed that arcs (C) and (D) are detected as a part. In this case, based on “extracted pattern a” and “extracted pattern c”, “@What THAILAND”, “@What FLOW”, “@Where DELTA”, “ @Where FLOW ”is generated.

The storage processing unit 35 of the meta information generating unit 32 adds the meta information generated as described above for the original sentence to the translated sentence of the original sentence and stores it in the parallel translation format DB 11.
Here, FIG. 5 will be described. FIG. 5 illustrates the data structure of the bilingual format DB 11.

  In the bilingual format DB 11, for each record, a record ID assigned to identify each record, the original text in the first language (here, Japanese), and the translated text in the second language (here, English translation) correspond to each other. Attached and stored. The storage processing unit 35 assigns a record ID starting from “0” to each record, reads the original sentence one by one from the original text DB 10, stores the original sentence, and further stores a translation for the original sentence. . This translated sentence is created by machine translation unit 12 machine-translating the original sentence read from original text DB 10, and storage processing unit 35 generates the original sentence by question type discriminating part 33 for the original sentence. The added meta information is added and stored in a predetermined record.

  In the fuzzy search performed by the fuzzy search unit 14, when an index is created for speeding up the search process, the index may be created when the creation is completed as described above. Note that when such an index is used, the file size of the database becomes large, and it is considered that the operable size is limited in terms of memory usage. However, in such a case, it is preferable to reduce the memory usage by taking measures such as dividing the database for each field to be searched.

Next, the search operation by the search device 1 in FIG. 1 will be described in more detail.
When the query sentence 2 is input to the search device 1, first, the semantic analysis unit 31 performs a semantic analysis of the query sentence 2 to create a conceptual structure of the query sentence 2. The semantic analysis unit 31 creates the conceptual structure of the query sentence 2 in the same manner as described above for creating the conceptual structure of the original sentence.

  FIG. 6 shows an example of creation of the conceptual structure of the query sentence 2 by the semantic analysis unit 31, and the user of the search apparatus 1 asks the query sentence 2 “Where is the prefecture with the highest production of potatoes?”. Is input and the search apparatus 1 is instructed to start search. Upon receiving the search start instruction, the search device 1 causes the semantic analysis unit 31 to perform semantic analysis of the query sentence 2, and as a result, the conceptual structure of the query sentence 2 illustrated in FIG. 6 is created.

  When the conceptual structure of the query sentence 2 is obtained, the query sentence question type discriminating unit 21 next selects the relationship between the concepts represented by the arcs in the conceptual structure of the query sentence 2 and the phrases connected by the arcs. Based on the part of speech, the type of the question expressed in the query sentence 2 is determined. Then, according to the determination result, meta information representing the type of the question is generated. A method of generating this meta information will be described with reference to FIG.

  The conceptual structure illustrated in FIG. 7 is the same as that in the example of FIG. Moreover, all of the extraction patterns a, b, c,..., N illustrated in FIG. 6 are the query sentence question type determination patterns 22 in FIG. The query sentence question type discrimination pattern 22 corresponds to the result of the semantic analysis on the arc in the conceptual structure of the query sentence 2, the part of speech of the phrase represented by the node connected by the arc, and the question type. It is a pattern attached. In particular, the query sentence question type determination pattern 22 is associated with the type of question formed by the question word when the part of speech of the phrase represented by the nodes connected by the arc is the question word. .

  For example, the “extracted pattern a” in FIG. 7 is that the result of the semantic analysis for the arc in the conceptual structure of the original sentence is “predicate”, and the phrase represented by the node connected by the arc is the place The question word "Where?" ] Represents a pattern. This “extracted pattern a” indicates that meta information “@Where” representing a question asking for a place among the types of 5W1H type questions is generated when this pattern matches.

  In this way, the query sentence question type determination unit 21 determines what the question represented by the query sentence 2 corresponds to among the 5W1H type questions by focusing on the question words included in the query sentence 2. To do. In this respect, a question type discriminating unit for original text 33 for discriminating which of the 5W1H type questions the question to which the word included in the original text is an answer corresponds to each word / phrase included in the original text, This is different from the query sentence question type determination unit 21.

  In the example of FIG. 7, the arc of (E) is detected as an arc whose result of semantic analysis is “predicate” from the conceptual structure, and this arc is the phrase “where? It is detected that it is connected to a node representing “”. Therefore, in this case, “@Where” is generated as meta information based on “extraction pattern a”.

Note that the query sentence question type determination unit 21 may include a central independent word in the generated meta information, similarly to the original sentence question type determination unit 33.
When the meta information is created by the query sentence question type discriminating unit 21, the search key is then created by the search key creating unit 13. The creation of this search key will be described with reference to FIG.

  First, the search key creation unit 13 obtains a translation of the query sentence 2 expressed in the first language into the second language from the machine translation unit 12, and uses this translation as a search key. In the example of FIG. 8, the machine translation unit 12 searches for the English translation “Where is prefecture where lot of production of potato exists?” For the query sentence 2 in Japanese, “Where is the prefecture with the highest production of potatoes?” This indicates that the key creation unit 13 acquires and uses it as a search key.

  Next, the search key creation unit 13 obtains the meta information representing the question type discrimination result by the query sentence 2 from the query sentence question type discrimination unit 21 and adds it to the translated sentence as the search key. This meta information is also used as a search key. In the example of FIG. 8, the search key creation unit 13 obtains the meta information “@Where” generated by the query statement question type determination unit 21 and representing the determination result of the question type by the query statement 2, and By adding it to the beginning of the English translation, this meta information is also used as a search key.

  Thereafter, the fuzzy search unit 14 uses the search key created by the search key creation unit 13 to perform a fuzzy search on the translation sentence in the second language in the parallel translation format DB 11, and corresponds to the translation sentence obtained as a result of this fuzzy search. The attached original text of the first language is extracted from the parallel translation format DB 11 and output. A specific example of this fuzzy search will be described with reference to FIG.

  As described above, in the translated language field of the second language of the bilingual form DB 11, the meta information generated by the source sentence question type determination unit 33 is added to the original sentence and stored. Therefore, the information desired to be obtained by performing an ambiguous search on the translated sentence field of the parallel translation DB 11 using the translated sentence of the query sentence 2 to which the meta information generated by the query sentence type discriminating unit 21 is added as a search key. Are obtained in the higher rank of the search result 3. In the example of FIG. 9, the meta information “@Where” is added to the English translation of the English field of the parallel translation DB 11 by using the English translation of the query sentence 2 with the meta information “@Where” added as a search key. Represents a hit in the higher rank in the result of the fuzzy search. That is, in the result of the fuzzy search, the meta information “@Where” matches as a character string, and the sentence containing “location” information is hit in the higher rank. Then, the Japanese original sentence about the English translation sentence obtained as a result of the fuzzy search is output from the search device 1 as the search result 3 according to this order.

  Since the search operation of the search device 1 in FIG. 1 is performed as described above, the information to be obtained is obtained in the higher rank of the search result 3, and furthermore, by the ambiguous search using the translated sentence of the original sentence and the query sentence 2 In addition, a decrease in search accuracy due to fluctuations in word level notation and word order is suppressed.

Next, FIG. 10 will be described. FIG. 10 shows an example of a hardware configuration of the search device 1 of FIG.
In this configuration example, the search device 1 is configured by a computer 50. The computer 50 includes an MPU 51, ROM 52, RAM 53, hard disk device 54, input device 55, output device 56, interface device 57, and recording medium drive device 58. These components are connected via a bus line 59, and various data can be exchanged under the management of the MPU 51.

An MPU (Micro Processing Unit) 51 is an arithmetic processing unit that controls the operation of the entire computer 50.
A ROM (Read Only Memory) 52 is a read-only semiconductor memory in which a predetermined basic control program is recorded in advance. The MPU 51 can control the operation of each component of the computer 50 by reading out and executing the basic control program when the computer 50 is started. As the ROM 52, a memory such as a flash memory whose storage data is nonvolatile may be used.

  A RAM (Random Access Memory) 53 is a semiconductor memory that can be written and read at any time and used as a working storage area as needed when the MPU 51 executes various control programs.

  The hard disk device 54 is a storage device that stores various control programs executed by the MPU 51 and various data. The MPU 51 can perform various control processes by reading and executing a predetermined control program stored in the hard disk device 54. When the search device 1 is configured using the computer 50, the hard disk device 54 stores in advance the original text DB 10, the query sentence question type determination pattern 22, and the original sentence question type determination pattern 34 of FIG. To leave. When the search device 1 is configured using the computer 50, the hard disk device 54 is also used as the parallel translation format DB 11.

  The input device 55 is, for example, a keyboard device or a mouse device. For example, when operated by the user of the search device 1, the input device 55 acquires and acquires various information inputs from the user associated with the operation content. The input information is sent to the MPU 51. The query sentence 2 that is an input to the search device 1 is received by the input device 55, for example.

  The output device 56 is, for example, a liquid crystal display or a speaker, and generates synthesized speech and displays various texts and images according to output data sent from the MPU 51. The search result 3 that is the output of the search device 1 is displayed on the output device 56, for example, or is output by reading out the search result 3 using synthesized speech.

  The interface device 57 manages the exchange of various information with various devices connected to the computer 50. The query statement 2 that is an input to the search device 1 may be output from another device, for example, and received by the interface device 57 directly or via a communication network such as the Internet. The search result that is the output of the search device 1 may be output from the interface device 57 and sent to another device directly or via a communication network such as the Internet. Further, an external storage device (not shown) connected to the interface device 57 directly or via a communication network such as the Internet may be used as a substitute or auxiliary for the hard disk device 54 provided in the computer 50 itself. Also good.

  The recording medium driving device 58 is a device that reads various control programs and data recorded on the portable recording medium 60. The MPU 51 may perform various control processes by reading and executing a predetermined control program recorded on the portable recording medium 60 via the recording medium driving device 58. As the portable recording medium 60, for example, a flash memory equipped with a CD-ROM (Compact Disc Read Only Memory), a DVD-ROM (Digital Versatile Disc Read Only Memory), or a USB (Universal Serial Bus) standard connector. and so on.

  In order to configure the search device 1 using such a computer 50, for example, a control program for causing the MPU 51 to perform various control processes described later is created. The created control program is stored in advance in the hard disk device 54 or the portable recording medium 60. Further, the original text DB 10, the query sentence question type determination pattern 22, and the original sentence question type determination pattern 34 are stored in the hard disk device 54 in advance. Then, a predetermined instruction is given to the MPU 51 to read and execute this control program. By doing so, the computer 50 functions as the parallel translation format DB 11, the machine translation unit 12, the search key creation unit 13, the fuzzy search unit 14, the query sentence question type determination unit 21, and the meta information generation unit 32. .

Next, various control processes performed by the MPU 51 of the computer 50 will be described. In the following description, the first language is Japanese and the second language is English.
First, FIG. 11 will be described. FIG. 11 is a flowchart showing the processing contents of the first example of the bilingual format DB creation processing. This process is a process of associating a Japanese original sentence stored in the original text DB 10 and its English translation sentence in the parallel translation format DB 11. However, in this first example, it is a process in the case where the generation of the meta information indicating the type of the question to which the word / phrase included in the original sentence is an answer and the storage into the parallel translation format DB 11 are not performed.

In FIG. 11, first, in S101, the MPU 51 performs a process of substituting the initial value “0” for the variable i.
Next, in S102, the MPU 51 performs a process of referring to the original text DB 10 to determine whether or not an original record having the value of the variable i at the time of execution of the process exists in the original text DB 10. In the present embodiment, it is assumed that the original text DB 10 stores a record ID starting from “0” and a Japanese original sentence, which are assigned to identify each record. .

  In the determination process of S102, when the MPU 51 determines that the original text record having the value of the variable i at the time of execution of the process exists in the original text DB 10 (when the determination result is Yes), the MPU 51 performs the process in S103. Proceed. On the other hand, when the MPU 51 determines that there is no original text record in the original text DB 10 in which the value of the variable i at the time of this process execution is a record ID (when the determination result is No), this bilingual format The DB creation process is terminated.

  Next, in S103, the original text is read from the record of the original text DB 10 using the value of the variable i at the time of this processing as the record ID, and machine translation processing is executed on the read original text to create an English translation of the original text. The MPU 51 performs the processing to be performed. The MPU 51 that performs this process functions as the machine translation unit 12.

  Next, in S104, the value of the variable i at the time of executing this process is set as a record ID, and a record storing the original sentence read from the original text DB 10 and the created English translation sentence in each field in the process of S103 is created and the bilingual form DB 11 The MPU 51 performs the process of storing the data. The MPU 51 that performs this process functions as the machine translation unit 12. The MPU 51 that performs this process functions as the storage processing unit 35.

  Next, in S105, the MPU 51 performs a process of substituting the value obtained by adding “1” to the value of the variable i at the time of this process into the variable i, and then returns to S102 to perform the above-described process. Repeated.

  The above processing is the first example of the bilingual format DB creation processing. When the MPU 51 performs this processing, the Japanese original text stored in the source text DB 10 and the English translation text are associated with each other and translated. Stored in the format DB 11.

  Next, FIG. 12 will be described. FIG. 12 is a flowchart showing the processing contents of the first example of the search processing. In this process, an ambiguous search is performed on the bilingual form DB 11 created by executing the first example of the bilingual form DB creating process described above using the English translation of the input Japanese query sentence 2 as a search key. This is the process to be performed.

  In FIG. 12, first, in S151, the MPU 51 executes a process of acquiring the query sentence 2 input to the search device 1 and executing a machine translation process on the acquired query sentence 2 to create an English translation of the query sentence 2. Do. The MPU 51 that performs this process functions as the machine translation unit 12.

  Next, in S152, the MPU 51 performs an ambiguous search on the English translation text field of the parallel translation DB 11 using the English translation text created by the processing of S151 as a search key. In the subsequent S153, the MPU 51 performs a process of reading the original text stored in the same record as the English translation obtained as a result of the fuzzy search from the parallel translation DB 11 and outputting it as the search result 3, and thereafter, this search is performed. The process ends. The MPU 51 that performs the processes of S152 and S153 functions as the fuzzy search unit 14.

  The above processing is the first example of the search processing, and the MPU 51 performs this processing, so that the search device 1 performs an ambiguous search, and the word expression in Japanese and the query sentence 2 in Japanese and the fluctuation of the word order. Is absorbed, and a decrease in search accuracy due to these fluctuations is suppressed.

  Next, FIGS. 13A and 13B will be described. 13A and 13B are flowcharts showing the processing contents of the second example of the bilingual format DB creation processing. This process is a process of associating a Japanese original sentence stored in the original text DB 10 and its English translation sentence in the parallel translation format DB 11. Further, in the second example, generation of meta information indicating the type of question in which the phrase included in the original sentence is an answer and storage in the bilingual form DB 11 are also performed.

First, in S201 of FIG. 13A, the MPU 51 performs a process of substituting the initial value “0” for the variable i.
Next, in S202, the MPU 51 performs a process of referring to the original text DB 10 to determine whether or not an original record having the value of the variable i at the time of execution of the process exists in the original text DB 10. Also in this embodiment, the original text DB 10 stores a record ID starting from “0” and a Japanese original sentence, which are assigned to each record to identify the record. To do.

  In the determination process of S202, when the MPU 51 determines that the original text record having the value of the variable i at the time of execution of the process exists in the original text DB 10 (when the determination result is Yes), the MPU 51 performs the process in S203. Proceed. On the other hand, when the MPU 51 determines that there is no original text record in the original text DB 10 in which the value of the variable i at the time of this process execution is a record ID (when the determination result is No), this bilingual format The DB creation process is terminated.

  Next, in S203, the original text is read from the record of the original text DB 10 using the value of the variable i at the time of execution of the process as the record ID, and the semantic analysis is performed on the read original text to create a conceptual structure for the original text. Processing is performed by the MPU 51. The MPU 51 that performs this process functions as the semantic analysis unit 31.

  Next, in S204, based on the result of the semantic analysis of the original text in the process of S203, and using the original text question type discrimination pattern 34, the above-described tag information about the words included in the English translation of the original text Is generated by the MPU 51. Details of the processing in S204 will be described later with reference to FIG. 13B. The MPU 51 that performs this process functions as the original text question type determination unit 33.

  Next, in S205, the original text is read from the record of the original text DB 10 using the value of the variable i at the time of execution of the process as a record ID, and machine translation processing is executed on the read original text to create an English translation of the original text. The MPU 51 performs the processing to be performed. The MPU 51 that performs this process functions as the machine translation unit 12.

  Next, in S206, a process of creating a record based on the original sentence read from the original text DB 10 in the process of S203 and the created English translation sentence with the tag information generated in the process of S204 and storing it in the bilingual form DB 11 Is performed by the MPU 51. Note that the value of the variable i at the time of executing this process is added to the created record as a record ID. The MPU 51 that performs this process functions as the storage processing unit 35.

  Next, in S207, the MPU 51 performs a process of substituting the value obtained by adding “1” to the value of the variable i at the time of this process into the variable i, and then returns to S202 to perform the above-described process. Repeated.

Next, details of the processing in S204 of FIG. 13A will be described using the flowchart of FIG. 13B.
First, in S211, the MPU 51 performs a process of substituting an initial value “0” for the variable j and initializing meta information and substituting empty data (null data).

  Next, in S212, the MPU 51 performs a process of referring to the original sentence question type determination pattern 34 to determine whether or not the pattern corresponding to the value of the variable j exists in the original sentence question type determination pattern 34. . In this embodiment, it is assumed that the original text question type determination pattern 34 is stored in order from the 0th in a predetermined storage area.

  In the determination process of S212, when the MPU 51 determines that an order pattern corresponding to the value of the variable j exists in the original question type determination pattern 34 (when the determination result is Yes), the MPU 51 advances the process to S213. On the other hand, when the MPU 51 determines that there is no pattern in the order corresponding to the value of the variable j in the original text question type determination pattern 34 (when the determination result is No), the process of FIG. 13B is performed. And the process returns to FIG. 13A.

  Next, in S213, whether or not the original sentence question type determination pattern 34 in the order corresponding to the value of the variable j matches the whole or part of the conceptual structure of the original sentence created in the process of S203 in FIG. 13A. The MPU 51 performs a process for determining the above as described with reference to FIG. When the MPU 51 determines that they match (when the determination result is Yes), the MPU 51 proceeds to S214. When it determines that they do not match (when the determination result is No), the MPU 51 proceeds to S215. Proceed with the process.

  Next, in S214, the meta information associated with the original text question type determination pattern 34 determined to be matched by the determination processing in S213 is further linked to the meta information created after the processing in FIG. 13B is started. Then, the MPU 51 performs processing to be added.

  Next, in S215, the MPU 51 performs a process of substituting the value obtained by adding “1” to the value of the variable j at the time of this process to the variable j. Thereafter, the process returns to S212, and the process described above is performed. Repeated.

  The above processing is the second example of the bilingual format DB creation processing. When the MPU 51 performs this processing, the Japanese original text stored in the source text DB 10 and the English translation text are associated with each other and translated. Stored in the format DB 11. In addition, by performing this processing by the MPU 51, generation of meta information indicating the type of question in which the phrase included in the original sentence is an answer and storage in the bilingual form DB 11 are also performed.

  Next, FIGS. 14A, 14B, and 14C will be described. FIG. 14A, FIG. 14B, and FIG. 14C are flowcharts showing the processing contents of the second example of the search processing. This process is performed for the bilingual form DB 11 created by executing the bilingual form DB creating process described above, and the English translation sentence for the input Japanese query sentence 2 and the question type of the query sentence 2 Is a process for performing an ambiguous search by using as a search key.

  First, in S251 of FIG. 14A, the MPU 51 performs a process of creating a search key from the query sentence 2 input to the search device 1. Details of the processing in S204 will be described later with reference to FIGS. 14B and 14C.

  Next, in S252, the MPU 51 performs an ambiguous search for the English translation field of the bilingual form DB11 using the search key created in the process of S251. In the subsequent S253, the MPU 51 performs a process of reading the original text stored in the same record as the English translation obtained as a result of the fuzzy search from the parallel translation DB 11 and outputting it as the search result 3, and thereafter this search is performed. End the process. The MPU 51 that performs the processing of S252 and S253 functions as the fuzzy search unit 14.

Next, details of the processing of S251 of FIG. 14A will be described using the flowchart of FIG. 14B.
First, in S261, the MPU 51 performs a process of acquiring the query sentence 2 input to the search device 1, performing semantic analysis on the acquired query sentence 2, and creating a conceptual structure for the query sentence 2. The MPU 51 that performs this process functions as the semantic analysis unit 31.

  Next, in S262, based on the result of the semantic analysis of the query sentence 2 in the process of S261, and using the query sentence question type determination pattern 22, the phrase included in the English translation of the query sentence 2 is determined. The MPU 51 performs the process for generating the tag information. Details of the processing in S262 will be described later with reference to FIG. 14C. The MPU 51 that performs this process functions as the query sentence question type determination unit 21.

  Next, in S263, the MPU 51 performs a process of executing a machine translation process on the query sentence 2 acquired by the process of S261 to create an English translation of the query sentence 2. The MPU 51 that performs this process functions as the machine translation unit 12. In this embodiment, in this machine translation process, the English translation sentence of the query sentence 2 is created by generating an English translation sentence corresponding to the conceptual structure of the query sentence 2 created by the process of S261. .

  Next, in S264, the MPU 51 performs a process of creating a search key by adding the tag information generated by the process of S262 to the English translation sentence created by the process of S263, and thereafter the process of FIG. 14B. And the process returns to FIG. 14A. The MPU 51 that performs this process functions as the search key creation unit 13.

Next, details of the processing of S262 of FIG. 14B will be described using the flowchart of FIG. 14C.
First, in S271 of FIG. 14C, the MPU 51 performs a process of substituting the initial value “0” for the variable k and initializing meta information and substituting empty data (null data).

  Next, in S272, the MPU 51 determines whether or not the query sentence question type determination pattern 22 has an order pattern corresponding to the value of the variable k with reference to the query sentence question type determination pattern 22. Do. In this embodiment, it is assumed that the query sentence question type determination pattern 22 is stored in order from the 0th in a predetermined storage area.

  In the determination process of S272, when the MPU 51 determines that the pattern in the order corresponding to the value of the variable k exists in the query sentence question type determination pattern 22 (when the determination result is Yes), the MPU 51 advances the process to S273. On the other hand, when the MPU 51 determines that the pattern of the order corresponding to the value of the variable k does not exist in the query sentence question type determination pattern 22 (when the determination result is No), the MPU 51 of FIG. The process ends and the process returns to FIG. 14B.

  Next, in S273, whether or not the query sentence question type determination pattern 22 in the order corresponding to the value of the variable k matches all or part of the conceptual structure of the query sentence 2 created in S261 of FIG. 14B. The MPU 51 performs the process of determining whether or not as described with reference to FIG. When the MPU 51 determines that they match (when the determination result is Yes), the MPU 51 proceeds to S274, and when it determines that they do not match (when the determination result is No), the MPU 51 proceeds to S275. Proceed with the process.

  Next, in S274, the meta information associated with the query sentence question type determination pattern 22 determined to match the meta information created after the processing of FIG. The MPU 51 performs the process of connecting and adding.

  Next, in S275, the MPU 51 performs a process of substituting the value obtained by adding “1” to the value of the variable k at the time of this process into the variable k, and then returns to S272 to perform the above-described process. Repeated.

  The above processing is the second example of the search processing. When the MPU 51 performs this processing, an ambiguous search is performed by the search device 1, and the word expression in Japanese and the query order in the original sentence and the query sentence 2 are fluctuated. Is absorbed, and a decrease in search accuracy due to these fluctuations is suppressed. In addition, since the fuzzy search is performed using the meta information indicating the type of the query of the query sentence 2 and the translated sentence of the query sentence 2 as a search key, the original sentence including the information to be known is more accurately obtained. The search result 3 is obtained.

The following additional notes are further disclosed with respect to the embodiment including the above examples.
(Appendix 1)
A bilingual format database in which the text of the first language to be searched and the translation of the text in the second language are stored in association with each other;
A machine translation unit for machine-translating the input query sentence in the first language into the second language;
A fuzzy search is performed using the query sentence translated into the second language as a search key with respect to the translated sentence stored in the bilingual format database, and the translated sentence is associated with the translated sentence as a result of the fuzzy search. An ambiguous search unit that extracts and outputs the text in the first language from the parallel translation database;
A search device comprising:
(Appendix 2)
In the bilingual format database, question type information indicating a type of question in which a phrase included in the text is an answer is further stored in association with the translation of the text,
Further comprising a discriminator for discriminating the type of question expressed by the query statement;
The fuzzy search unit performs the fuzzy search using the query type determined by the determination unit in addition to the query sentence translated into the second language, as the search key;
The search device according to supplementary note 1, wherein:
(Appendix 3)
A semantic analysis unit that performs semantic analysis of the text;
Based on the result of the semantic analysis, the question type information for generating the question type information for the phrase included in the translated sentence for the text and storing it in the parallel translation format database in association with the translated sentence for the text A generator,
Further comprising
The fuzzy search unit performs the fuzzy search on the translated sentence and the question type information associated with the translated sentence stored in the parallel translation format database.
The search device according to supplementary note 2, wherein:
(Appendix 4)
The question type information generation unit generates meta information including the question type information and a phrase included in a translated sentence for the text that is an answer to a question of a type represented by the question type information, Attached to the translation of the text and stored in the bilingual database,
The fuzzy search unit performs the fuzzy search on a translation sentence that is stored in the parallel translation format database and to which the meta information is added.
The search device according to Supplementary Note 3, wherein
(Appendix 5)
The semantic analysis unit analyzes the meaning of a syntax tree created by parsing the text, and connects each node with a node representing the phrase that constitutes the text. Generating a conceptual structure of the text composed of arcs representing the relationship of the concepts;
The question type information generation unit extracts an arc representing a concept relationship previously associated with the question type from the conceptual structure of the text, and the type of question associated with the extracted arc Is generated as the question type information about the words and phrases represented by the nodes connected by the extracted arc,
The search device according to supplementary note 3 or 4, characterized in that:
(Appendix 6)
The semantic analysis unit further performs semantic analysis of the query statement,
The machine translation unit performs machine translation of the query sentence into the second language by generating a sentence of the second language corresponding to a result of semantic analysis of the query sentence.
The search device according to any one of supplementary notes 3 to 5, characterized in that:
(Appendix 7)
The discriminator connects each node and each node representing each word constituting the query sentence by analyzing the meaning of a syntax tree created by parsing the query sentence. To generate a conceptual structure of the query statement composed of arcs representing the concept relationship between words, and connected by the arc to the conceptual relationship represented by the arc in the conceptual structure of the query statement The search device according to any one of appendices 2 to 6, wherein the type of the question expressed by the query sentence is determined based on the part of speech of the phrase represented by the node.
(Appendix 8)
The search according to claim 7, wherein the determination unit determines the type of the question expressed by the query sentence based on the type of the question word for the part of speech of the word that is a question word. apparatus.
(Appendix 9)
9. The search device according to any one of appendices 2 to 8, wherein the question type is a 5W1H type in a 5W1H type question.
(Appendix 10)
A program for causing a computer to execute a search process for a bilingual format database in which a text in a first language to be searched and a translated sentence in the second language for the text are stored in association with each other,
Machine-translated the input query sentence in the first language into the second language;
An ambiguous search is performed using a query sentence translated into the second language as a search key with respect to the translated sentence stored in the parallel translation format database,
Extracting and outputting the text in the first language associated with the translated sentence as a result of the fuzzy search from the bilingual format database;
A program for causing the computer to execute processing.
(Appendix 11)
In the bilingual format database, question type information indicating a type of question in which a phrase included in the text is an answer is further stored in association with the translation of the text,
The program further causes the computer to execute a process of determining a type of a question expressed by the query statement,
In the process of performing the fuzzy search, the fuzzy search is further performed using the type of the question determined by the determination process in addition to the query sentence translated into the second language as the search key.
The program according to appendix 10.
(Appendix 12)
The program is
Perform semantic analysis of the text,
Based on the result of the semantic analysis, generate the question type information for the phrase included in the translated sentence for the text, and store it in the parallel translation format database in association with the translated sentence for the text.
Causing the computer to further perform processing;
In the process of performing the fuzzy search, the fuzzy search is performed on the translated sentence and the question type information associated with the translated sentence stored in the parallel translation format database.
The program according to appendix 11.
(Appendix 13)
In the process of generating the question type information, meta information including the question type information and a phrase included in a translated sentence for the text that is an answer to the type of question represented by the question type information is generated. , Added to the translation of the text and stored in the bilingual format database,
In the process of performing the fuzzy search, the fuzzy search is performed on the translation sentence stored in the bilingual format database and having the meta information added thereto.
The program according to attachment 12.
(Appendix 14)
In the process of performing the semantic analysis, by analyzing the meaning of the syntax tree created by parsing the text, the phrase representing the phrase constituting the text is connected to each node by the phrase Generating a conceptual structure of the text composed of arcs representing the relationship between the concepts of each other;
In the process of generating the question type information, an arc representing a concept relationship previously associated with the question type is extracted from the conceptual structure of the text, and the question associated with the extracted arc Generating information as the question type information about the words and phrases represented by the nodes connected by the extracted arc,
The program according to appendix 12 or 13.
(Appendix 15)
The program further causes the computer to execute processing for performing semantic analysis of the query statement,
In the machine translation process, machine translation of the query sentence into the second language is performed by generating a sentence in the second language corresponding to a result of semantic analysis of the query sentence.
The program according to any one of appendices 12 to 15.
(Appendix 16)
In the process of determining the type of the question, by analyzing the meaning of the syntax tree created by parsing the query sentence, each node representing each word and phrase constituting the query sentence and each Generating a conceptual structure of the query statement composed of arcs representing the relationship of concepts between words by connecting nodes, and the relationship of concepts represented by arcs in the conceptual structure of the query statement; The program according to any one of supplementary notes 11 to 15, wherein the type of a question expressed by the query sentence is determined based on a part of speech expressed by a node connected by an arc.
(Appendix 17)
The program according to appendix 16, wherein, in the process of determining the type of the question, the type of the question expressed by the query sentence is determined based on the type of the interrogative word when the part of speech of the phrase is a questionable word. .
(Appendix 18)
The program according to any one of Supplementary Notes 11 to 17, wherein the question type is a 5W1H type in a 5W1H type question.
(Appendix 19)
A search method for a bilingual format database in which a text in a first language to be searched and a translation in a second language for the text are stored in association with each other,
Machine-translated the input query sentence in the first language into the second language;
An ambiguous search is performed using a query sentence translated into the second language as a search key with respect to the translated sentence stored in the parallel translation format database,
Extracting and outputting the text in the first language associated with the translated sentence as a result of the fuzzy search from the bilingual format database;
A search method characterized by that.
(Appendix 20)
In the bilingual format database, question type information indicating a type of question in which a phrase included in the text is an answer is further stored in association with the translation of the text,
Determining the type of question represented by the query statement;
In the fuzzy search, in addition to the query sentence translated into the second language, the fuzzy search is further performed using the type of the question determined by the determination as the search key.
The search method according to supplementary note 19, characterized by:

1 Search Device 2 Query Statement 3 Search Result 10 Original Text DB
11 Bilingual DB
DESCRIPTION OF SYMBOLS 12 Machine translation part 13 Search key creation part 14 Fuzzy search part 21 Query sentence question classification judgment part 22 Query sentence question classification judgment pattern 31 Semantic analysis part 32 Meta information generation part 33 Original sentence question classification judgment part 34 Original sentence question classification Discrimination pattern 35 Storage processing unit 40 Translation dictionary 50 Computer 51 MPU
52 ROM
53 RAM
54 hard disk device 55 input device 56 output device 57 interface device 58 recording medium drive device 59 bus line 60 portable recording medium

Claims (10)

The text of the first language to be searched, the translated text of the text in the second language, and the question type information indicating the type of question to which the phrase included in the text is an answer are stored in association with each other A bilingual database,
A machine translation unit for machine-translating the input query sentence in the first language into the second language;
A discriminator for discriminating the type of the question expressed by the query statement;
An ambiguous search is performed on the translated sentence stored in the bilingual form database using a query sentence translated into the second language and a question type determined by the determination unit as a search key. An ambiguous search unit that extracts and outputs text in a first language associated with a translated sentence that is a search result from the bilingual format database;
A search device comprising: A semantic analysis unit that performs semantic analysis of the text;
Based on the result of the semantic analysis, the question type information for generating the question type information for the phrase included in the translated sentence for the text and storing it in the parallel translation format database in association with the translated sentence for the text A generator,
Further comprising
The fuzzy search unit performs the fuzzy search on the translated sentence and the question type information associated with the translated sentence stored in the parallel translation format database.
The search device according to claim 1 . The question type information generation unit generates meta information including the question type information and a phrase included in a translated sentence for the text that is an answer to a question of a type represented by the question type information, Attached to the translation of the text and stored in the bilingual database,
The fuzzy search unit performs the fuzzy search on a translation sentence that is stored in the parallel translation format database and to which the meta information is added.
The search device according to claim 2 . The semantic analysis unit analyzes the meaning of a syntax tree created by parsing the text, and connects each node with a node representing the phrase that constitutes the text. Generating a conceptual structure of the text composed of arcs representing the relationship of the concepts;
The question type information generation unit extracts an arc representing a concept relationship previously associated with the question type from the conceptual structure of the text, and the type of question associated with the extracted arc Is generated as the question type information about the words and phrases represented by the nodes connected by the extracted arc,
The search device according to claim 2 or 3 , wherein The semantic analysis unit further performs semantic analysis of the query statement,
The machine translation unit performs machine translation of the query sentence into the second language by generating a sentence of the second language corresponding to a result of semantic analysis of the query sentence.
The search device according to any one of claims 2 to 4 , characterized in that: The discriminator connects each node and each node representing each word constituting the query sentence by analyzing the meaning of a syntax tree created by parsing the query sentence. To generate a conceptual structure of the query statement composed of arcs representing the concept relationship between words, and connected by the arc to the conceptual relationship represented by the arc in the conceptual structure of the query statement based on the word parts of speech which are nodes represent the search device according to any one of claims 1-5, characterized in that to determine the type of questions that are represented by the query statement . The determination section, for those the word part of speech is interrogative, according to claim 6, characterized in that to determine the type of questions that are represented by the query statement based on the type of該疑Wheels Search device. The search device according to any one of claims 1 to 7 , wherein the information indicating the type of question is a type of 5W1H in a 5W1H type question. The text of the first language to be searched, the translated text of the text in the second language, and the question type information indicating the type of question to which the phrase included in the text is an answer are stored in association with each other A program for causing a computer to execute a search process for a parallel translation database,
Machine-translated the input query sentence in the first language into the second language;
Determining the type of question represented by the query statement;
An ambiguous search is performed using the query sentence translated into the second language for the translation sentence stored in the bilingual form database and the type of the determined question as a search key,
Extracting and outputting the text in the first language associated with the translated sentence as a result of the fuzzy search from the bilingual format database;
A program for causing the computer to execute processing. The text of the first language to be searched, the translated text of the text in the second language, and the question type information indicating the type of question to which the phrase included in the text is an answer are stored in association with each other a search method performed by the computer to pair with and bilingual format database,
The computer is
Machine-translated the input query sentence in the first language into the second language;
Determining the type of question represented by the query statement;
An ambiguous search is performed using the query sentence translated into the second language for the translation sentence stored in the bilingual form database and the type of the determined question as a search key,
Extracting and outputting the text in the first language associated with the translated sentence as a result of the fuzzy search from the bilingual format database;
A search method characterized by that.

Images